search

facebookresearch / vissl

VISSL is FAIR's library of extensible, modular and scalable components for SOTA Self-Supervised Learning with images.
https://vissl.ai
MIT License
3.25k stars 332 forks source link

How to use fairness indicators for SEER models? #544

Open Prabharoop opened 2 years ago

Prabharoop commented 2 years ago

❓ How to use fairness indicators for SEER models?

Hi,

First of all, thank you for providing vissl, it's a great tool. I would like to know what changes have to be made to use the metrics provided in the fairness indicator project for the SEER RegNet models (for eg. ; https://github.com/facebookresearch/vissl/blob/main/projects/fairness_indicators/label_associations_indicator1.md).

Inputs: I have created a sample set from the casual conversations dataset and added that to the vissl system as mentioned. Outputs: I expect the fairness metrics as an output as mentioned in the sample documentation.

I am using google colab, with a single GPU.

Fairness indicators: https://github.com/facebookresearch/vissl/tree/main/projects/fairness_indicators SEER models: https://github.com/facebookresearch/vissl/tree/main/projects/SEER#pretrained-models-weights

QuentinDuval commented 2 years ago

Hi @Prabharoop,

Thanks a lot for your interest in VISSL and our fairness indicators :) I am not the most qualified for this part, Priya should be able to best explain it.

@prigoyal Could you please take a look at this issue?

Thank you, Quentin